Compare apples to apples. Waymo drives on limited number of streets that it knows. Maybe google tesla fsd beta and watch a few videos of how the software behaves and judge for yourself. From what I’ve seen you get a cool graphic of what the car software “sees”, with different colors.
Like this, for instance ?
https://www.reddit.com/r/softwaregore/comments/jg1nfb/interesting_tesla_stoplight_detection/
1 Like
Maybe Tesla has a secret deal with Migros and that’s why it interprets Coop logo as a stop light.
But seriously, we’re talking about FSD beta here, available to select drivers in USA, and this footage is clearly from Switzerland. So high lol value, low relevance.
1 Like
There it goes, with Tesla being “the best”… Yeah no.
https://www.reuters.com/article/us-autos-selfdriving-safety-idUKKBN27D1B7
1 Like
The critical difference in the Super Cruise system is a driver-facing infrared camera to make sure he or she is paying attention to the road and is ready to take over manual control when necessary, said Kelly Funkhouser, head of connected and automated vehicle testing at Consumer Reports.
How can less autonomy be better scored? I’m already annoyed by the EU regulation that one needs to touch the steering wheel every 60 seconds.
but falling short on its ability to maintain a driver’s focus on the road.
That’s the whole fucking point of autonomous driving…
2 Likes
Nobody claimed that Tesla’s autopilot was the best in the market. This test is anyway worthless if they didn’t test the FSD beta. You only can offer uninformed bias. I have a feeling you will be proved wrong once the new update goes live.
Videos of FSD beta start being published on Youtube.
Example:
While this is certainly impressive, there are still some worrying points.
My notes on this 7 minutes footage:
- 00:30 : the car does not realize that the lane is one-way so it does not want to switch to the left lane ( this one is just a minor annoyance. i think it is actually good that in case of doubt it stays on the right). Intervention of the driver needed.
- 02:02 : the car is confused by a construction site and does not turn. Driver intervention needed.
- 03:30 : while turning right at a crossroad, the car does not see a median strip. Driver intervention needed or it would have clipped the car.
- 05:14 : the car detects correctly that a truck is not parked correctly. Nice!
- 06:28 : while going across a crossroad, the car does not see that another car is coming from the right. Driver intervention needed, this would have been an accident.
Overall, on this 7 minutes video there were two major interventions needed and two minor ones. While Tesla’s autonomous driving is impressive 98% of the time, I still feel like the remaining “edge cases” are not ready to go live and may take time to correct. Let’s see how they improve on these.
4 Likes
I would love to watch some of these videos, but I will only have time after holidays. Ok, 7 min is a short time for so much to happen.
I’ve been driving in southern Italy and here I wish good luck to any self driving car. Lanes are washed out, including stop lines. Drivers keep driving between the lanes, or switch lanes without blinking or looking. I had to brake hard twice when I was about to overtake such a wobbly driver. I used the horn, but I would be more eager to punch them in the face. They also overtake you on the right if you maintain a safe distance to the car in front, while driving the left lane. And on top of that, there are massive holes which you have to slalom around or your car won’t survive too long.
But for Switzerland this tech could be ready pretty soon!
Btw I read one or two reaction to fsd beta, and what impressed me was that in an intersection with poor visibility, the Tesla was creeping forward slowly, just like a human would.
1 Like
This is really bad 
I mean, is this what should be driving people around within a year ? Not 5 minutes driving without an accident and they are driving on empty roads, in a checkboard plan, in good weather conditions.
Can you imagine these cars driving in Europe ? Even in Switzerland, the roads are far more challenging and packed with cars, and I am not even speaking about the weather.
I am really sad because I would love having self-driving cars. What a disappointment.
Self-driving cars will not be on the market within 1 year for sure.
Maybe check what the state of the art is, I think it’s better than the tesla beta (quite a few companies have decent disengagement stats).
But certainly don’t listen to this guy 
Here’s why:
the reports (…) should not be used to “compare one company with another or reach broad conclusions on technological capabilities
Sure, but there’s a difference between those and two minor + two majors interventions in 7 minutes 
Did I read it here or elsewhere? The owner of Teslas are them training the AI for tesla. It’s not that you can use that thing to read a book while driving. You are just helping Musk making his billions
Btw. why they don’t mount some top cameras as well?
Ok I watched that video and another one. It is pretty underwhelming that it will basically crash if you give it a few minutes. But I also read the comments. Every disengagement like this is sent back to Tesla and trains the neural network, so it’s crucial that more and more beta testers give it a go. I am already impressed by stop light and stop sign behavior.
I find it fantastic that they provide this 3d visualisation that you can even rotate. I mean, come on! Which of the current production cars has this option? Even as a beta? None. I dislike that the rendered image is a bit wobbly. The lines are shaking even if the car is standing. Intuition tells me, the image should be stabilizing with time, not keep shaking.
I’m pretty sure this “beta” will still take a few years until they are at 99.999% cases. Right now it doesn’t seem they are even at 99.0%
The AI doesn’t have a concept of standing still. It draws what it “sees”, either cars or lines. They should start to “divide and conquer”-up the way they reason. Like our brain does.
I don’t think me or you are qualified to say what this particular AI can or cannot do, but it should certainly possible for an AI to figure out that it is not moving and that this particular object is in this particular location, and not jump around a few centimeters with each frame.
I’d like to counter with these two videos:
Almost 30mins w/o much interaction from the driver needed.
I guess we’re all victims of our confirmation biases
By the way, Elon is in his top Twitter form.
I cringe at this kind of humor, I wonder what his co-workers and big shareholders feel. Are they already immune to this or do they also shake their heads in disbelief.
1 Like
So anyone buying TSLA stocks? I’m waiting for the SpaceX IPO 
1 Like